Support __array__ and __list__ protocols

Inputs to bokeh graphing functions seem to be able to take one of the following forms

DataFrame, dict, OrderedDict, list, tuple, ndarray

I wonder if we could extend this to include anything that supports the __array__ or __iter__ protocols. In my own experiments I wrote this function

def normalize_data(data):
    """ Normalize input data to supported types """
    if isinstance(data, (np.ndarray, pd.DataFrame, list, tuple, dict,
        OrderedDict)):
        return data
    if np and hasattr(data, '__array__'):
        result = np.array(data)
        # TODO: Other parts of Bokeh should support numpy record dtypes
        #       convert to DataFrame for now
        if isinstance(result.dtype.descr, list):  # record dtype
            return pd.DataFrame(result)
        else:
            return result
    if isinstance(data, Iterable):
        return list(data)

    raise TypeError("Input type %s not supported" % type(data).__name__)

and inserted it into bokeh/charts/_builder.py:_adapt_values. This allowed me to use Blaze expressions directly in Bokeh plots which felt pretty cool.

More generally would it make sense to normalize inputs even further, perhaps using only a single data type within Bokeh, presumably either a DataFrame or OrderedDict. I notice that some functions seem to branch on the type of _values.

[bryevdv]

That is really cool! we should enable this right after 0.8 release I think. Regarding a single type, we are doing out best not to have a hard dependency on Pandas, but also not to require unnecessary copying. Our DataAdapter class is a (not yet complete) stab at a very thin wrapper that presents a “pandas-like” interface, regardless of that the actual data is.

[bryevdv]

closing, bokeh.charts is gone. Possibly makes sense to suggest to HV